我的实验有什么问题(试图预测汽车销量)？

Question

我有这样的数据集(只是其中的一个示例)：

DATE_REF,MONTH,YEAR,DAY_OF_YEAR,DAY_OF_MONTH,WEEK_DAY,WEEK_DAY_1,WEEK_DAY_2,WEEK_DAY_3,WEEK_DAY_4,WEEK_DAY_5,WEEK_DAY_6,WEEK_DAY_7,WEEK_NUMBER_IN_MONTH,WEEKEND,WORK_DAY,AMOUNT_SOLD
20100101,1,2010,1,1,6,0,0,0,0,0,1,0,1,0,0,0
20100102,1,2010,2,2,7,0,0,0,0,0,0,1,1,1,0,2
20100103,1,2010,3,3,1,1,0,0,0,0,0,0,2,1,0,0
20100104,1,2010,4,4,2,0,1,0,0,0,0,0,2,0,1,12830
20100105,1,2010,5,5,3,0,0,1,0,0,0,0,2,0,1,19200
20100106,1,2010,6,6,4,0,0,0,1,0,0,0,2,0,1,22930
20100107,1,2010,7,7,5,0,0,0,0,1,0,0,2,0,1,23495
20100108,1,2010,8,8,6,0,0,0,0,0,1,0,2,0,1,23215
20100109,1,2010,9,9,7,0,0,0,0,0,0,1,2,1,0,172
20100110,1,2010,10,10,1,1,0,0,0,0,0,0,3,1,0,0
20100111,1,2010,11,11,2,0,1,0,0,0,0,0,3,0,1,18815
20100112,1,2010,12,12,3,0,0,1,0,0,0,0,3,0,1,25415
20100113,1,2010,13,13,4,0,0,0,1,0,0,0,3,0,1,25262
20100114,1,2010,14,14,5,0,0,0,0,1,0,0,3,0,1,27967
20100115,1,2010,15,15,6,0,0,0,0,0,1,0,3,0,1,26352
20100116,1,2010,16,16,7,0,0,0,0,0,0,1,3,1,0,202
20100117,1,2010,17,17,1,1,0,0,0,0,0,0,4,1,0,10
20100118,1,2010,18,18,2,0,1,0,0,0,0,0,4,0,1,20295
20100119,1,2010,19,19,3,0,0,1,0,0,0,0,4,0,1,25982
20100120,1,2010,20,20,4,0,0,0,1,0,0,0,4,0,1,24745
20100121,1,2010,21,21,5,0,0,0,0,1,0,0,4,0,1,28087
20100122,1,2010,22,22,6,0,0,0,0,0,1,0,4,0,1,28417
20100123,1,2010,23,23,7,0,0,0,0,0,0,1,4,1,0,115
20100124,1,2010,24,24,1,1,0,0,0,0,0,0,5,1,0,5
20100125,1,2010,25,25,2,0,1,0,0,0,0,0,5,0,1,20185
20100126,1,2010,26,26,3,0,0,1,0,0,0,0,5,0,1,25932
20100127,1,2010,27,27,4,0,0,0,1,0,0,0,5,0,1,31710
20100128,1,2010,28,28,5,0,0,0,0,1,0,0,5,0,1,21020
20100129,1,2010,29,29,6,0,0,0,0,0,1,0,5,0,1,51460
20100130,1,2010,30,30,7,0,0,0,0,0,0,1,5,1,0,670
20100131,1,2010,31,31,1,1,0,0,0,0,0,0,6,1,0,17

我正在使用Azure上的以下实验来预测新日期的AMOUNT_SOLD (DATE_REF)：

​

​

然后我部署了Web并测试了预测，但是我得到的只是AMOUNT_SOLD列的0。

我可能错过了什么？

Answer 1

虽然我很想复制你的Azure ML实验，但我没有足够的数据。但我所做的如下：

​

​

我复制了示例数据，然后将其乘以4次(添加行x2)。然后分割数据 (70%/30%)，随机种子7(可复制结果)。增强的决策树回归具有默认参数。在调优模型超参数上，我选择了AMOUNT_SOLD作为标签列。然后对评分模型、评分模型和评分模型进行评价。

​

​

准确度/决定系数相当好。

在此之后，要将其部署为web服务，必须首先从培训实验中设置一个预测实验。Setup Web Service > Predictive Experiment你的实验将像魔法一样移动。

​

​

默认情况下， experiment输入模块将放在实验的顶部。I移动它并连接到Score Model的右侧，这样当您输入web服务的参数时，就会使用您经过训练的模型来预测它。

在得分模型模块之后，我在Dataset模块中放置了一个选择列，并且只选择了名为评分标签的列。本专栏包含模型的预测。然后，我使用编辑元数据模块重命名得分标签列，然后将其传递给Web 模块。

您的实验现在可以部署为web服务了。

为了预测新值，我使用当前日期详细信息作为输入来测试web服务。(，尽管DATE_REF输入必须是20170818 :D )

​

​

然后输出结果如下：

​

​

您的web服务现在可以预测新的值。